Gray scale image watermarking using fuzzy entropy and Lagrangian twin SVR in DCT domain

In this paper, the effect of low, middle and high frequency DCT coefficients are investigated onto gray scale image watermarking in terms of imperceptibility and robustness. The performance of Lagrangian twin support vector regression (LTSVR), which was successfully applied on synthetic datasets obtained from UCI repository for various kinds of regression problems by Balasundaram et al. [9], onto image watermarking problem, is validated by embedding and extracting the watermark on different standard and real world images. Also the good learning capability of image characteristics provides the good imperceptibility of the watermark and robustness against several kinds of image processing attacks verifies the high generalization performance of LTSVR. Through the experimental results, it is observed that significant amount of imperceptibility and robustness is achieved using low frequency (LF) DCT coefficients as compared to middle frequency (MF) and high frequency (HF) DCT coefficients as well as state-of-art technique.